P10.09 Cluster analysis of Bovine TB using multiple methods, and considerations when constructing a Composite Index

Abstract

Multiple methods were employed to analyze spatial clustering of Bovine Tuberculosis (TB) breakdowns in cattle across the High Risk Area (HRA) of England using data from 2013-2021. Located in South-West England and the West Midlands region, the HRA is one of three TB Risk Areas in England, and is often associated with the majority of TB incidents in the country (APHA, 2022). These methods employed Spatial Statistics tools and Space Time Pattern Mining tools in ArcGIS Pro, use of SaTScan^tm, and ETL processing using FME and Model Builder.

The first method built on techniques applied to data from the Edge Area of England by Downs et al. (2021), utilizing circular 25km² spatial units and set cut-off criteria for three rolling 7-year periods to determine the strength of evidence for a local reservoir. The circular spatial units aligned to a grid of centroids 2km apart, creating overlap between the spatial units, and addressing potential uncertainty as to the boundary of any reservoir found with contiguous spatial units.

The second method utilized the same circular 25km² spatial units but this time estimated TB infection using herd incidence per 100 herd-years at risk. Initially this was calculated per circle, and later applied the nearest 100 herds to each circle.

The third method used Hot Spot Analysis via the Getis-Ord Gi* statistic to identify statistically significant hot and cold spots of TB per year across contiguous hexagonal 25km² spatial units. These same spatial and temporal units were used in Cluster and Outlier Analysis using the Anselin Local Moran’s I statistic as the fourth method. TB breakdown hot spots and cold spots were identified, as well as statistically significant spatial outliers. For both the Getis-Ord Gi* and Anselin Local Moran’s I statistics, parameters were tested using ArcGIS’s Conceptualization of Spatial Relationships to optimize cluster identification.

The fifth method involved Space Time Pattern Mining and implemented the Getis-Ord Gi* and Anselin Local Moran’s I statistics across a space-time cube of comparable spatial and temporal extent.

The sixth method utilized SaTScan^tm to analyze the space-time clustering of TB breakdowns for five biennial periods within the time of the study. Clusters were identified and then ranked by their statistical significance for each period.

Spatial and spatio-temporal correlation and difference were noted across the different methods, and Composite Indexing was investigated for its viability in combining the multiple methods into a single index. The methodology for integration of spatial and temporal units and extents is described, noting the importance of expert opinion and degree of subjectivity in parameter selection when creating an index.